Machine learning for software analysis

Basics for use

The use of machine learning is being researched and successfully applied in many areas to answer complex questions, for example in facial recognition or text analysis. There is also a strong need for the use of AI methods in IT itself, such as the evaluation or error analysis of software, which currently requires a lot of manual work. However, ML algorithms usually work on input data of a fixed structure and size. Software source code, on the other hand, is hierarchically structured and of variable length.
In a master’s thesis in collaboration with itestra GmbH and the Technical University of Munich, a method was developed to transform source code into vectors so that it can be used as input for ML algorithms. To do this, a transformation is first carried out in graph form and then an alignment is performed, which in turn uses a supervised learning approach.
The method was tested as part of a larger research project to evaluate software quality.